The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads core library assets, styles, and plugin-specific scripts from well-known CDNs including jsDelivr (cdn.jsdelivr.net) and Unpkg (unpkg.com). These are used to load dependencies like D3.js, KaTeX, and Highlight.js for browser-side rendering.
[DATA_EXFILTRATION]: The markmap frontmatter configuration supports extraJs and extraCss fields. These allow users to specify arbitrary JavaScript and CSS to be included and executed within the generated HTML file. If the agent processes a malicious Markdown file, this feature could be abused for Cross-Site Scripting (XSS) or to exfiltrate information from the environment where the mindmap is viewed.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests untrusted Markdown data. Maliciously crafted Markdown content (such as acrostics or hidden instructions in comments like ) could attempt to influence the agent's behavior during the transformation process.
Ingestion points: Transformer.transform() processes raw Markdown content from files or standard input.
Boundary markers: The skill uses specific Markdown-it plugins and YAML frontmatter delimiters, but does not implement explicit safety filters for embedded instructions.
Capability inventory: Access to the local filesystem via CLI, ability to start a web server (markmap --dev), and dynamic script loading in the browser context.
Sanitization: The skill uses markdown-it and cheerio for parsing, but the extraJs feature intentionally allows raw script injection.
[COMMAND_EXECUTION]: The markmap-cli utility provides several shell commands for generating mindmaps and running a development server. The server uses chokidar for file watching and Hono for handling HTTP requests, which involves reading local files and serving them over the network.

markmap